A Guide to Digital Systems and Measurements

A Guide to Digital TelevisionSystems and MeasurementsLast active picture sample First active picture sampleCOMPOSITE: Full analog line digitizedCopyright ® 1997, Tektronix, Inc. All rights reserved.A Guide to Digital Television Systems and MeasurementsAcknowledgmentAuthored by David K. Fibushwith contributions from:Bob ElkindKenneth AinsworthSome text and figures used in this booklet reprint-ed with permission from Designing Digital Sys-tems published by Grass Valley Products.iiiiiiare covered in this booklet: 1.A definition of digitaltelevision including digi-tal basics, video stan-dards, conversionsbetween video signalforms, and digital audioformats and standards. 2.Considerations in select-ing transmission systempassive components suchas cable, connectors andrelated impedance match-ing concerns are dis-cussed as well as contin-ued use of the passiveloop-through concept.Much of the transmissionsystem methods and hard-ware already in place maybe used for serial digital.3.New and old systemsissues that are particularlyimportant for digital videoare discussed, includingequalization, reclocking,and system timing.serial digital signals canbe characterized in threedifferent aspects: usage,methods, and operationalenvironment. Types ofusage include: designer’sworkbench, manufactur-ing quality assurance,user equipment evalua-tion, system installationand acceptance testing,system and equipmentmaintenance, and perhaps most important, opera-tion.5.Several measurementmethods are covered indetail, including serialwaveform measurements,definition and detectionof errors, jitter effects andmeasurements, and sys-tem testing with specialtest signals.measurement are required todetermine the health of thedigital video transmissionsystem and the signals it car-ries. Analysis methods forthe program video and audiosignals are well known.However, measurements forthe serial form of the digi-tized signal differ greatlyfrom those used to analyzethe familiar baseband videosignal.This guide addresses manytopics of interest to designersand users of digital televisionsystems. You’ll find informa-tion on embedded audio(Section 3), system timingissues (Section 4), and sys-tem timing measurements(Section 6). Readers who arealready familiar with digitaltelevision basics may wish tostart with Section 4, SystemHardware and Issues. Discus-sion of specific measurementtopics begins in Section 6.page 1page 2page 3width of 36 MHz. So, data compression is nothing new; digital just makes it easier. For composite NTSC signals, there are additional gamut considerations when con-verting from the color differ-ence domain. NTSC trans-mitters will not allow 100 percent color amplitude on signals with high luminance level (such as yellow). This is because the transmitter carrier will go to zero for sig-nals greater than about 15 percent above 1 volt. Hence, there’s a lower gamut limit for some color difference sig-nals to be converted to NTSC for RF transmission.As part of the encoding pro-cess, sync and burst are added as shown in Figure2-3. Burst provides a refer-ence for decoding back to components. The phase of burst with respect to sync edge is called SCH phase, which must be carefully con-trolled in most studio appli-cations. For NTSC, 7.5 IRE units of setup is added to the luminance signal. This poses some conversion difficulties,particularly when decodingback to component. Theproblem is that it’s relativelyeasy to add setup; but remov-ing it when the amplitudesand timing of setup in thecomposite domain are notwell controlled can result inblack level errors and/or sig-nal disturbances at the endsof the active line.Sampling and QuantizingThe first step in the digitiz-ing process is to “sample”the continuous variations ofthe analog signal as shown inFigure 2-4. By looking at theanalog signal at discrete timeintervals, a sequence of volt-age samples can be stored,manipulated and later,reconstructed.In order to recover the analogsignal accurately, the samplerate must be rapid enough toavoid missing important in-formation. Generally, thisrequires the sampling fre-quency to be at least twicethe highest analog frequency.In the real world, the fre-quency is a little higher thantwice. (The Nyquist Sam-pling Theorem says that theinterval between successivesamples must be equal to orless than one-half of theperiod of the highest fre-quency present in the signal.)The second step in digitizingvideo is to “quantize” byassigning a digital number tothe voltage levels of the sam-pled analog signal – 256 lev-els for 8-bit video, 1024 for10-bit video, and up to sev-eral thousand for audio.To get the full benefit of digi-tal, 10-bit processing isrequired. While most currenttape machines are 8-bit,SMPTE 125M calls out 10-bitas the interface standard.Processing at less than 10bits may cause truncationand rounding artifacts, par-ticularly in electronicallygenerated pictures. Visibledefects will be revealed inthe picture if the quantiza-tion levels are too coarse (toofew levels). These defectscan appear as “contouring”of the picture. However, thegood news is that randomnoise and picture details pre-sent in most live video sig-nals actually help concealthese contouring effects byadding a natural randomnessto them. Sometimes the num-ber of quantizing levels mustbe reduced; for example,when the output of a 10-bitprocessing device feeds an8-bit recorder. In this case,contouring effects are mini-mized by deliberately addinga small amount of randomnoise (dither) to the signal.This technique is known asrandomized rounding.Digital Video StandardsAlthough early experimentswith digital technology werebased on sampling the com-posite (NTSC or PAL) signal,it was realized that for thehighest quality operation,component processing wasnecessary. The first digitalstandards were component.Interest in composite digitalwas revived when Ampexand Sony announced a com-posite digital recording for-mat, which became known aspage 4signals C B and C R, which are scaled versions of the signals B-Y and R-Y.The sampling structure defined is known as “4:2:2.”This nomenclature is derived from the days when multi-ples of NTSC subcarrier were being considered for the sam-pling frequency. This ap-proach was abandoned, but the use of “4” to represent the luminance sampling fre-quency was retained. The task force mentioned above examined luminance sam-pling frequencies from12MHz to 14.3 MHz. They selected 13.5 MHz as a com-promise because the sub-multiple 2.25 MHz is a factor common to both 525- and 625-line system. Some extended definition TV sys-tems use a higher resolution format called 8:4:4 which has twice the bandwidth of 4:2:2. Parallel component digital. Rec. 601 described the sam-pling of the signal. Electrical interfaces for the data pro-duced by this sampling were standardized separately by SMPTE and the EBU. The parallel interface for525/59.94 was defined by SMPTE as SMPTE Standard 125M (a revision of the ear-lier RP-125), and for 625/50 by EBU Tech 3267 (a revi-sion of the earlier EBU Tech 3246). Both of these were adopted by CCIR and are included in Recommenda-tion 656, the document defining the hardware inter-face.The parallel interface uses eleven twisted pairs and 25-pin “D” connectors. (The early documents specified slide locks on the connec-tors; later revisions changed the retention mechanism to 4/40 screws.) This interface multiplexes the data words in the sequence C B, Y, C R, Y, C B…, resulting in a data rate of 27 Mwords/s. The timing sequences SAV and EAV were added to each line to represent Start of Active Video and End of Active Video. The digital active line contains 720 luminance sam-ples, and includes space for representing analog blanking within the active line. Rec. 601 specified eight bits of precision for the data words representing the vid-eo. At the time the standards were written, some partici-pants suggested that this may not be adequate, and provi-sion was made to expand the interface to 10-bit precision. Ten-bit operation has indeed proved beneficial in many circumstances, and the latest revisions of the interface standard provide for a 10-bit interface, even if only eight bits are used. Digital-to-ana-log conversion range is cho-sen to provide headroom above peak white and foot-room below black as shown in Figure 2-5. Quantizing lev-els for black and white are selected so the 8-bit levels with two “0”s added will have the same values as the 10-bit levels. Values 000to 003 and 3FF to3FC are reserved for synchronizingD-2. Initially, these machineswere projected as analogin/out devices for use inexisting analog NTSC andPAL environments; digitalinputs and outputs were pro-vided for machine-to-ma-chine dubs. However, thepost-production communityrecognized that greateradvantage could be taken ofthe multi-generation capabil-ity of these machines if theywere used in an all digitalenvironment.Rec. 601.RecommendationITU-R BT.601 (formerly CCIRRecommendation 601) is nota video interface standard,but a sampling standard. Rec.601 evolved out of a jointSMPTE/EBU task force todetermine the parameters fordigital component video forthe 525/59.94 and 625/50television systems. Thiswork culminated in a seriesof tests sponsored by SMPTEin 1981, and resulted in thewell known CCIR Recom-mendation 601. This docu-ment specified the samplingmechanism to be used forboth 525 and 625 line sig-nals. It specified orthogonalsampling at 13.5MHz forluminance, and 6.75MHzfor the two color differencepage 5page 6purposes. Similar factorsdetermine the quantizing val-ues for color difference sig-nals as shown in Figure 2-6.Figure 2-7 shows the location of samples and digital words with respect to an analog horizontal line. Because the timing information is carried by EAV and SAV, there is no need for conventional syn-chronizing signals, and the horizontal intervals (and the active line periods during the vertical interval) may be used to carry ancillary data.The most obvious applica-tion for this data space is to carry digital audio, and doc-uments are being prepared by SMPTE to standardize the format and distribution of the audio data packets.Rec. 601/656 is a well proven technology with a full range of equipment available for production and post produc-tion. Generally, the parallel interface has been super-seded by a serial implemen-tation, which is far more practical in larger installa-tions. Rec. 601 provides all of the advantages of both dig-ital and component opera-tion. It’s the system of choice for the highest possible qual-ity in 525- or 625-line systems.Parallel composite digital.The composite video signal is sampled at four times the (NTSC or PAL) subcarrier frequency, giving nominal sampling rates of 14.3 MHz for NTSC and 17.7 MHz for PAL. The interface is stan-dardized for NTSC as SMPTE 244M and, at the time of writing, EBU documentation is in process for PAL. Both interfaces specify ten-bit pre-cision, although D-2 and D-3machines record only eight bits to tape. Quantizing of the NTSC signal (shown in Figure 2-8) is defined with a modest amount of headroom above 100% bars, a small footroom below sync tip and the same excluded values as for component.PAL composite digital has been defined to minimizecomposite. Also, the full excursion of the composite signal has to be represented by 256 levels on 8-bit recorders. Nevertheless, com-posite digital provides a more powerful environment than analog for NTSC and PAL installations and is a very cost effective solution for many users.As with component digital, the parallel composite inter-face uses a multi-pair cable and 25-pin “D” connectors. Again, this has proved to be satisfactory for small and medium sized installations, but practical implementation of a large system requires a serial interface.16:9 widescreen component digital signals.Many televi-sion markets around the world are seeing the intro-duction of delivery systems for widescreen (16:9 aspect ratio) pictures. Some of these systems, such as MUSE in Japan and the future ATVsystem in the USA, are intended for high-definition images. Others, such as PALplus in Europe will use 525- or 625-line pictures. Domestic receivers with 16:9 displays have been intro-duced in many markets, and penetration will likely increase significantly over the next few years. For most broadcasters, this creates the need for 16:9 program material.Two approaches have been proposed for the digital rep-resentation of 16:9 525- and 625-line video. The first method retains the sampling frequencies of the Rec. 601 standard for 4:3 images (13.5MHz for luminance). This “stretches” the pixel represented by each data word by a factor of 1.33 hori-zontally, and results in a loss of 25 percent of the horizon-tal spatial resolution when compared to Rec. 601 pic-tures. For some applications, this method still provides ad-equate resolution, and it has the big advantage that muchpage 7existing Rec. 601 equipment can be used.A second method maintains spatial resolution where pix-els (and data words) are added to represent the addi-tional width of the image. This approach results in 960 luminance samples for the digital active line (compared to 720 samples for a 4:3 pic-ture, and to 1920 samples for a 16:9 high definition pic-ture). The resulting sampling rate for luminance is18MHz. This system pro-vides the same spatial resolu-tion as Rec. 601 4:3 images; but existing equipment designed only for 13.5 MHz cannot be used. The addi-tional resolution of this method may be advantageous when some quality overhead is desired for post produc-tion or when up-conversion to a high-definition system is contemplated.SMPTE 267M provides for both the 13.5MHz and18MHz systems. Because 16:9 signals using 13.5MHzsampling are electricallyindistinguishable from 4:3signals, they can be con-veyed by the SMPTE 259Mserial interface at 270Mb/s.It’s planned that SMPTE259M (discussed below) willbe revised to provide forserial transmission of18MHz sampled signals,using the same algorithm, ata data rate of 360Mb/s.Serial Digital VideoParallel connection of digitalequipment is practical onlyfor relatively small installa-tions, and there’s a clearneed for transmission over asingle coaxial cable. This isnot simple as the data rate ishigh, and if the signal weretransmitted serially withoutmodification, reliable recov-ery would be very difficult.The serial signal must bemodified prior to transmis-sion to ensure that there aresufficient edges for reliableclock recovery, to minimizethe low-frequency content ofthe transmitted signal, and tospread the transmittedenergy spectrum so thatradio-frequency emissionproblems are minimized.In the early 1980s, a serialinterface for Rec. 601 signalswas recommended by theEBU. This interface used 8/9block coding and resulted ina bit rate of 243 Mb/s. Itsinterface did not support ten-bit precision signals, andthere were some difficultiesin producing reliable, costeffective, integrated circuits.The block coding based inter-face was abandoned and hasbeen replaced by an interfacewith channel coding thatuses scrambling and conver-sion to NRZI. The serial inter-face has been standardized asSMPTE 259M and EBU Tech.3267, and is defined for bothcomponent and compositesignals including embeddeddigital audio.Conceptually, the serial digitalinterface is much like a carriersystem for studio applica-tions. Baseband audio andvideo signals are digitized andcombined on the serial digital“carrier” as shown in Figure2-11. (It’s not strictly a carriersystem in that it’s a basebanddigital signal, not a signalmodulated on a carrier.) Thebit rate (carrier frequency) isdetermined by the clock rateof the digital data: 143Mb/sfor NTSC, 177Mb/s for PALand 270Mb/s for Rec. 601component digital. Thewidescreen (16:9) componentsystem defined in SMPTE 267will produce a bit rate of360Mb/s.Parallel data representing thesamples of the analog signalis processed as shown in Fig-ure 2-12 to create the serialdigital data stream. The par-allel clock is used to loadsample data into a shift regis-ter, and a times-ten multipleof the parallel clock shiftsthe bits out, LSB (least signif-icant bit) first, for each 10-bitdata word. If only 8 bits ofdata are available at theinput, the serializer placespage 8zeros in the two LSBs to complete the 10-bit word. Component signals do not need further processing as the SAV and EAV signals on the parallel interface provide unique sequences that can be identified in the serial domain to permit word fram-ing. If ancillary data such as audio has been inserted into the parallel signal, this data will be carried by the serial interface. The serial interface may be used with normal video coaxial cable.The conversion from parallel to serial for composite signals is somewhat more complex. As mentioned above, the SAV and EAV signals on the parallel component interface provide unique sequences that can be identified in the serial domain. The parallel composite interface does not have such signals, so it is necessary to insert a suitable timing reference signal (TRS) into the parallel signal before serialization. A diagram of the serial digital NTSC hori-zontal interval is shown in Figure 2-13. The horizontal interval for serial digital PAL would be similar except that the sample location is slightly different on each line, building up two extra samples per field. A three-word TRS is inserted in the sync tip to enable word fram-ing at the serial receiver, which should also remove the TRS from the received serial signal.The composite parallel inter-face does not provide for transmission of ancillary data, and the transmission of sync and burst means that less room is available for insertion of data. Upon con-version from parallel to serial, the sync tips may be used. However, the data space in NTSC is sufficient for four channels of AES/EBU digital audio. Ancillary data such as audio may be added prior to serialization, andthis would normally be per-formed by the same co-pro-cessor that inserts the TRS. Following the serialization of the parallel information the data stream is scrambled by a mathematical algorithm then encoded into NRZI (non-return to zero inverted) by a concatenation of the follow-ing two functions:G1(X) = X9+ X4+ 1G2(X) = X + 1At the receiver the inverse of this algorithm is used in the deserializer to recover the correct data. In the serial dig-ital transmission system, the clock is contained in the data as opposed to the parallel system where there’s a sepa-rate clock line. By scram-bling the data, an abundance of transitions is assured as required for clock recovery. The mathematics of scram-bling and descrambling lead to some specialized test sig-nals for serial digital systems that are discussed later in this guide.Encoding into NRZI makes the serial data stream polar-ity insensitive. NRZ (non return to zero, an old digital data tape term) is the familiar circuit board logic level, high is a “1” and low is a “0.” For a transmission system it’s convenient to not require a certain polarity of the signal at the receiver. As shown in Figure 2-14, a data transition is used to represent each “1”page 9and there’s no transition for a data “0.” The result’s that it’s only necessary to detect tran-sitions; this means either polarity of the signal may be used. Another result of NRZI encoding is that a signal of all “1”s now produces a tran-sition every clock interval and results in a square wave at one-half the clock fre-quency. However, “0”s pro-duce no transition, which leads to the need for scram-bling. At the receiver, the ris-ing edge of a square wave at the clock frequency would be used for data detection.Rate Conversion – Format ConversionWhen moving between com-ponent digital and composite digital in either direction, there are two steps: the actual encoding or decoding, and the conversion of the sample rate from one stan-dard to the other. The digital sample rates for these two formats are different:13.5MHz for component dig-ital and 14.3 MHz for NTSC composite digital (17.7 MHz for PAL). This second step is called “rate conversion.”Often the term rate conver-sion is used to mean both encoding/decoding and resampling of digital rates.Strictly speaking, rate con-version is taking one samplerate and making anothersample rate out of it. For ourpurposes, we’ll use the term“format conversion” to meanboth the encode/decode stepand resampling of digitalrates. The format conversionsequence depends on direc-tion. For component to com-posite, the usual sequence israte conversion followed byencoding. For composite tocomponent, the sequence isdecoding followed by rateconversion. See Figure 2-15.It’s much easier to do pro-duction in componentbecause it isn’t necessary towait for every color framingboundary (four fields forNTSC and eight for PAL) tomatch two pieces of video to-gether; instead, they can bematched every two fields(motion frame). This pro-vides four times the opportu-nity. Additionally, compo-nent is a higher quality for-mat since luminance andchrominance are handledseparately. To the extent pos-sible, origination for compo-nent environment produc-tion should be accomplishedcomponent. A high qualitycomposite-to-component for-mat converter provides a rea-sonable alternative.After post-production work,component digital oftenneeds to be converted tocomposite digital. Sourcesthat are component digitalmay be converted for inputto a composite digitalswitcher or effects machine.Or a source from a compo-nent digital telecine may beconverted for input to a com-posite digital suite. Further-more, tapes produced incomponent digital may needto be distributed or archivedin composite digital.The two major contributorsto the quality of this processare the encoding or decodingprocess and the sample rateconversion. If either one isfaulty, the quality of the finalproduct suffers. In order toaccurately change the digitalsample rate, computationsmust be made between twodifferent sample rates and in-terpolations must be calcu-lated between the physicallocation of the source pixeldata and the physical loca-tion of destination pixel data.Of the 709,379 pixel loca-tions in a PAL compositedigital frame, all except onemust be mapped (calculated).In order to accomplish thiscomputationally intense con-version, an extremely accu-rate algorithm must be used.If the algorithm is accurateenough to deal with PALconversions, NTSC conver-sions using the same algo-rithm are simple. To ensurequality video, a sophisticatedalgorithm must be employedand the hardware must pro-duce precise coefficients andminimize rounding errors.page 10page 11generally more economical to maintain separate audio thus eliminating the need for numerous mux (multiplexer) and demux (demultiplexer) modules. In developing the SDI standard, SMPTE 259M, one of the key considerations was the ability to embed at least four channels of audioin the composite serial digi-tal signal which has limitedancillary data space. A basicform of embedded audio forcomposite video is docu-mented in SMPTE 259M andthere’s a significant amountof equipment in the fieldusing this method for bothcomposite and componentsignals. As engineers wereusing the SMPTE 259M spec-ifications for their designs, itbecame clear that a moredefinitive document wasneeded. A draft proposedstandard is being developedfor embedded audio whichincludes such things as dis-tribution of audio sampleswithin the available compos-ite digital ancillary dataspace, the ability to carry 24-bit audio, methods to handlenon-synchronous clockingand clock frequencies otherthan 48 kHz, and specifica-tions for embedding audio incomponent digital video.Ancillary data -posite digital video allowsancillary data only in itsserial form and then only inthe tips of synchronizing sig-nals. Ancillary data is notcarried in parallel compositedigital as parallel data is sim-ply a digital representation ofthe analog signal. Figures3-2 and 3-3 show the ancil-lary data space available incomposite digital signalswhich includes horizontalsync tip, vertical broadpulses, and vertical equaliz-ing pulses.A small amount of the hori-zontal sync tip space isreserved for TRS-ID(timingreference signal and identifi-cation) which is required fordigital word framing in thedeserialization process. Sincecomposite digital ancillarydata is only available in theserial domain, it consists of10-bit words, whereas bothcomposite and componentparallel interconnects may beeither 8 or 10 bits. A sum-mary of composite digitalancillary data space for NTSCis shown in Table 3-1. Thirtyframes/second will provide9.87Mb/second of ancillarydata which is enough for fourchannels of embedded audiowith a little capacity left over.PAL has slightly more ancil-lary data space due to thehigher sample rate; howeverpage 12page 13HANC (horizontal ancillary data) and VANC (vertical ancillary data) with the SMPTE defining the use of HANC and the EBU defining the use of VANC. Word lengths are 10-bit for HANC specifically so it can carry embedded audio in the same format as for composite digi-tal and 8-bit for VANCspecifically so it can be car-ried by D-1 VTRs (on those vertical interval lines that are recorded). Total data space is shown in Table 3-2. Up to 16channels of embedded audio are specified for HANC in the proposed standard and there’s room for significantly more data.Not all of the VANC data space is available. For in-stance, the luminance sam-ples on one line per field are reserved for DVITC (digital vertical interval time code)and the chrominance sam-ples on that line may bedevoted to video index. Also,it would be wise to avoid using the vertical interval switch line and perhaps, the subsequent line where data might be lost due to relock after the switch occurs.Ancillary data formatting.Ancillary data is formatted into packets prior to multi-plexing it into the video data stream as shown in Figure 3-5. Each data block may contain up to 255 user data words provided there’s enough total data space available to include five(composite) or seven (compo-nent) words of overhead. For composite digital, only the vertical sync (broad) pulses have enough room for the full 255 words. Multiple data packets may be placed in individual ancillary data spaces, thus providing a rather flexible data commu-nications channel.At the beginning of each data packet is a header usingword values that are exclud-ed for digital video data and reserved for synchronizing purposes. For composite video, a single header word of 3FCh is used whereas for component video, a three-word header 000h 3FFh 3FFh is used. Each type of data packet is identified with a different Data ID word.Several different Data ID words are defined to orga-nize the various data packets used for embedded audio.The Data Block Number(DBN) is an optional counter that can be used to provide sequential order to ancillary data packets allowing a receiver to determine ifembedded audio for digital PAL is rarely used.There’s considerably more ancillary data space available in component digital video as shown in Figure 3-4. All of the horizontal and vertical blanking intervals are avail-able except for the small amount required for EAV (end of video) and SAV (start of video) synchronizing words.The ancillary data space has been divided into two types。